i-PI + cp2k simulation crashes when using multiple keywords in FUNCTIONAL command in cp2k

[Luis]

Hello i-PI users,

I recently started using i-PI with cp2k and it has been working great in general. However, I recently ran into a problem (I think it may be a bug) with this section of the cp2k input file:

      &XC_FUNCTIONAL

        &LIBXC

          FUNCTIONAL MGGA_X_SCAN MGGA_C_SCAN

        &END LIBXC

      &END XC_FUNCTIONAL

The i-PI + cp2k simulation runs fine only when I use a single entry in the FUNCTIONAL keyword (e.g. FUNCTIONAL MGGA_X_SCAN). But it crashes when I use FUNCTIONAL MGGA_X_SCAN MGGA_C_SCAN. When I use just cp2k, the simulation runs fine either way (that's why I ended up posting this issue here and not in the cp2k group). 

The fact that one of the i-PI+cp2k simulations with almost identical settings (the only difference is the FUNCTIONAL line in the cp2k input file) runs fine makes me believe that it is not an issue of the specific system or the setup, that's why I  am not providing any more details of the input files.

I was wondering if this is a known issue and whether there is any way of working around it. 

For reference, I am using cp2k 3.0 and a development version of i-PI that I downloaded in April 2017.

Thanks for your time!

[Michele Ceriotti]

Hi Luis, 

I do not know well enough the implementation of meta-gga in CP2K to know whether there is something special one should take care of when using multiple FUNCTIONAL statements. i-PI uses a very high-level call to the force evaluator in CP2K so I think it would be best to get in touch with the CP2K developers. 

However, the real question is whether the code crashes (e.g. as in a memory leak) or if the calculation does not converge. In the latter case, a possible culprit is the fact that your i-PI input might be in the wrong units, and so some functionals still manage to converge, and others just never do.

Basically, you need to give more information if you want help.

Cheers

[Michele Ceriotti]

OK just to report that this ended up being a  - now fixed - bug in libxc. All well what ends well.